What is the central thesis of "ChatGPT is Bullshit," and why does the authors' choice of word matter?

Because a language model is indifferent to the truth of its output — designed to seem plausible, not to be accurate — its falsehoods are, in Frankfurt's exact sense, bullshit; naming it correctly changes how we handle it.

Hicks, Humphries and Slater (2024) apply Frankfurt's definition directly to LLMs. The fit is precise: an LLM is built to produce text that looks truth-apt, with no underlying concern for whether it's true — which is the very definition of bullshit. So its errors are neither lies (it isn't trying to deceive about facts) nor hallucinations (it isn't perceiving), but bullshit.

Why the word matters — terminology guides how people respond:

• "Hallucination" implies a mostly-truthful system that occasionally glitches → leads to misguided "fix the perception" patches.
• "Bullshit" implies a system indifferent to truth by design → the right warning is to distrust every output, even correct ones, because correctness is incidental.

The key consequence: "the inaccuracies show that it is bullshitting even when it's right." Getting a true answer doesn't mean the system was tracking truth.

Tip: The lesson for users: fluent, confident output is engineered to seem credible. Treat plausibility as zero evidence of accuracy and verify independently.